Evolution of atomic-scale roughening on Si(001)-(231) surfaces resulting
from high temperature oxidation
J. V. Seiple and J. P. Pelz
Department of Physics, The Ohio State University, Columbus, Ohio 43210
~Received 3 October 1994; accepted 30 January 1995!
Scanning tunneling microscopy was used to study surface morphology on Si~001!-~231! samples
following elevated temperature O2 exposure as a function of pressure ~131028,P ox,531026
Torr!, temperature ~500,T s ,700 °C!, and dose @10– 800 langmuir ~L!#. At low O2 doses ~D ox,50
L; T s >600 °C and P ox,131027 Torr! preferred B-type step retraction is observed, but single
A-domain formation is prevented due to step pinning by nucleated oxide clusters. These pinning
centers roughen the surface via step ‘‘fingering’’ at low doses, while at higher doses result in the
formation of three-dimensional conical islands, similar to oxidation-induced ‘‘growth features’’
reported previously by Smith and Ghidini @J. Electrochem. Soc. 129, 1300 ~1982!# at higher
temperatures. Surface etching rates were determined by measuring the island heights as a function
of exposure, and a sticking coefficient s of 0.04 ~60.02! was estimated for O2 reaction on Si~001!
at 600 °C. For moderate O2 doses ~,1000 L! surface etching was found to dominate SiO2 growth
up to pressures several orders of magnitude above the commonly accepted oxidation ‘‘critical line,’’
causing significant atomic-scale roughening under these oxidation conditions. © 1995 American
Vacuum Society.
I. INTRODUCTION
It is well known that exposing Si to O2 can, depending on
substrate temperature T s and oxygen pressure P ox , either
etch the surface ~via the evolution of volatile SiO! or lead to
the formation of a solid SiO2 film. A critical curve is often
referenced1,2 which separates the pressure–temperature conditions for which etching is the dominant process ~the socalled active oxidation regime! from those which produce
SiO2 growth ~the passive oxidation regime!. At high temperatures ~.1000 °C! micrometer-sized ‘‘growth’’ features
were observed at high doses within a narrow ‘‘transition region’’ between the active and passive regimes.2 Less is
known about how the atomic-scale Si surface morphology
depends on oxidation conditions, particularly at lower temperatures ~,800 °C! which are increasingly important to
modern semiconductor processing. Recently it has been
shown that atomic-scale surface roughening results on both
Si~001! @Ref. 3# and Si~111! @Ref. 4# surfaces over a range of
low temperature and low pressure conditions in which oxide
nucleation and surface etching occur simultaneously. To date,
however, relatively little is still known about the boundaries
of this ‘‘roughening’’ regime, or how the surface morphology
evolves with temperature, pressure, and O2 dose, particularly
on the technologically important Si~001! surface. Since
many Si processing steps involve transient, low-temperature
exposure to background levels of O2 and H2O, a knowledge
of the conditions which produce surface roughening and a
characterization of the roughness thus produced has great
practical importance.5
We have used scanning tunneling microscopy ~STM! to
study oxidation-induced roughening of Si~001! surfaces in
this low temperature roughening regime. The focus of the
present article is a description of how the surface roughness
evolves with O2 dose. The temperature dependence of oxide
cluster nucleation and atomic modeling of the nucleation
772
J. Vac. Sci. Technol. A 13(3), May/Jun 1995
process will be discussed in detail elsewhere.6 At low O2
doses ~D ox<50 langmuir ~L!, where 1 L[131026 Torr s! we
have observed the preferential retraction of ‘‘B-type’’ surface
steps, in agreement with ion-sputtering studies by Bedrossian
and Klitsner.7 However, the preferential retraction is not
complete due step pinning by nanometer-sized oxide
clusters.3 At higher doses, these pinning sites lead to the
creation of three-dimensional conical ‘‘island’’ structures as
prolonged surface etching removes the surrounding material.
These structures may be early stages of the micrometer-sized
‘‘growth’’ features observed by Smith and Ghidini2 at higher
temperatures, pressures, and doses. At T s >600 °C and
P ox>631028 Torr, the average island height and rms surface
roughness were found to scale approximately linearly with
dose. The surface etching was directly determined from the
dependence of island height versus dose, which indicated a
sticking coefficient s>0.04 ~60.02! for O2 on Si under
these conditions. Finally, we have started to delineate the
boundaries in temperature–pressure space that result in significant atomic-scale surface roughening, and have found
them to extend to pressures several orders of magnitude
above the critical line for net SiO2 growth.
II. EXPERIMENT
Experimental procedures for these experiments have been
discussed in detail elsewhere.8 In short, Si~001! samples ~25
3730.4 mm, n type, ,0.25° miscut toward @110#! were resistively ‘‘flashed’’ to 1250 °C to remove any oxide present.
Clean surfaces exhibited evenly spaced A-type steps ~upper
terrace dimer rows parallel to step! and B-type steps ~upper
terrace dimer rows perpendicular to step!.9 Figure 1~a! shows
a typical large-area scan ~300 nm3300 nm! of the clean
surface. Although this scan area is too large to show the
atomic rows, one can easily distinguish the smooth A-type
steps from the more ragged B-type steps. Once atomic-
0734-2101/95/13(3)/772/5/$6.00
©1995 American Vacuum Society
772
773
J. V. Seiple and J. P. Pelz: Evolution of atomic-scale roughening
FIG. 1. ~a! Top-view gray scale STM topograph of 3003300 nm2 area of the
clean Si~001!-~231! starting surface, miscut toward @110#. Smooth A-type
and rougher B-type steps are marked. ~b! 28328 nm2 closeup showing
typical defect density. A- and B-type terraces are labeled. ~c! 3003300 nm2
area after 50 L O2 dose at 231028 Torr and 600 °C, showing reduction in
area of B-type terraces. ~c! Ratio of A- to B-type terrace areas vs O2 dose
~see the text!.
resolution STM scans confirmed that the clean surface had a
low defect density @,2%, as shown in the close-up image in
Fig. 1~b!#, the sample was heated to the desired oxidation
temperature and then exposed to O2 . The oxidation pressure
was determined by the ion pump current which had been
previously calibrated against a standard ion gauge. All hot
filaments were off during an actual dose. Samples were
quenched to room temperature about 6 s after the O2 was
turned off and then scanned. Samples were reflashed before
each O2 exposure.
III. RESULTS
For low oxygen pressures and doses at 600 °C, we found
that oxidation-induced etching preferentially etches B-type
steps over A-type. Figure 1~a! shows a clean starting surface
with well defined, evenly spaced A- and B-type steps, while
Fig. 1~c! shows a different area of the surface after exposure
to 50 L of O2 at 231028 Torr and 600 °C. As discussed
elsewhere,3 oxidation under these conditions primarily results in surface etching through the formation of surface vacancies ~via desorption of SiO!. At these temperatures the
vacancies are mobile7 and diffuse to step edges causing them
to retract. Bedrossian and Klitsner7 have shown that diffusing vacancies created by ion sputtering at T s >450 °C preferentially terminate at B-type steps, causing them to preferentially retract to the point where a single A-domain surface
may be formed. Figure 1~c! shows that oxidation-induced
etching at 231028 Torr and 600 °C also leads to clear, preferential reaction of B-type steps. In contrast to sputter etching, however, the formation of long step fingers ~primarily on
B-type steps! prevents the formation of a single A-domain
surface. We have shown elsewhere3 that these step fingers
are caused by nanometer-sized structures ~probably small oxJVST A - Vacuum, Surfaces, and Films
773
ide clusters! which pin the steps during etching. As the dose
becomes higher, etching becomes prevalent at B-type kinks
on A-type steps, causing the local step orientation to randomize. At this point steps have a large degree of mixed A- and
B-type character, and hence both types of terraces become
equally etched.
To quantify this preferential etching behavior, we have
measured the relative areas of A-type terraces ~above an
A-type step! and B-type terraces ~above a B-type step!. A
terrace is assigned to be A type ~B type! if its dimer rows run
perpendicular ~parallel! to the original @110# sample miscut
direction. The ratio of A-terrace areas to B-terrace areas as a
function of dose is shown in Fig. 1~d!. The A:B ratio is
observed to increase from a value near one with increasing
dose ~indicating preferential retraction of B-type steps! up to
doses in the range of 30–50 L. Thereafter, the A:B ratio
decreased for higher doses as the step directions randomized.
We note here that Wurm et al.10 have used low energy electron microscopy ~LEEM! to study oxidation-induced etching
under similar conditions, but on Si~001! samples with much
lower miscut angle. On large terraces, they observed preferential etching in a direction parallel to the dimer rows, behavior consistent with the preferential retraction of B-type
steps reported here. One difference, however, is that their
LEEM measurements did not indicate substantial step pinning at 600 °C even at O2 pressures as high as 531028 Torr.
We will return to this point later.
As discussed in Ref. 3 ~and originally proposed in Ref.
11!, a pinning site during step retraction will first form a long
finger, but as step retraction proceeds the end of the finger
will eventually break away to form a monolayer-high island.
As subsequent steps pass by the pinning site, the island will
progressively increase in height, eventually leading to the
formation of a three-dimensional etch structure. We have
studied this island evolution with oxygen dose. Figure 2
shows a sequence of large-area STM scans after an O2 dose
D ox ranging from 100 to 800 L, at constant pressure 631028
Torr and temperature 600 °C. Each scan is shown both as a
top-view grey scale image and as a three-dimensional perspective. The grey scale images show that for all D ox the
islands are situated on a flat, vicinal base surface. Close-up
scans of areas between the islands ~not shown! reveal Si
dimer rows and surface steps, indicating that the base surface
is not covered with a continuous oxide film even up to
D ox5800 L. The perspective views show that the islands
grow in size with increasing D ox . Note that the vertical scale
~220→50 Å! in the perspective views has been greatly expanded relative to the horizontal scale to enhance island visibility.
Figure 3~a! shows how the island heights vary with D ox .
For each island, a ‘‘peak’’ height was measured relative to a
plane fit through the vicinal base surface, and then an average peak height was determined for all the islands in a given
scan. The error bars reflect scan-to-scan variations in the
measured average peak island height for a given dose. We
see from Fig. 3~a! that the average island height scales approximately linearly with dose, particularly for D ox.200 L.
As will be discussed elsewhere,6 few new islands were found
to nucleate for D ox.200 L; the existing islands simply grew
774
J. V. Seiple and J. P. Pelz: Evolution of atomic-scale roughening
774
FIG. 2. Top-view gray scale and three-dimensional perspective views of 3003300 nm2 topographs, after an O2 dose of ~a! 100 L, ~b! 200 L, ~c! 400 L, ~d! 800
L, respectively. The vertical scale in the perspective views extends from 220 to 150 Å. All oxidations performed at T s >600 °C and P ox>631028 Torr.
in size. Consequently, the average island height approached
the maximum island height at higher doses, allowing a direct
determination of the average vertical etching rate
R>0.002460.0006 ML/s for this temperature ~600 °C!
and pressure ~631028 Torr!. This in turn implies a sticking
coefficient for O2 reaction on Si~001! under these conditions
of s>0.0460.02. The large uncertainty in s is mainly due to
the large uncertainty ~650%! in our absolute pressure measurements. This value for s is within the range reported by
several recent studies,12,13 but is significantly higher than the
value14 of s reported by Wurm et al.10
Figure 3~b! shows how the measured rms surface roughness G varies with dose. At each dose, G was measured for
several 500 nm3500 nm scans ~the data shown in Fig. 2
were cut from these larger scans!, and the error bars reflect
FIG. 3. ~a! Plot of the average island height vs O2 dose. ~b! Plot of the rms
surface roughness G vs O2 dose.
J. Vac. Sci. Technol. A, Vol. 13, No. 3, May/Jun 1995
measured scan-to-scan variations. For our oxidation conditions, we see that G scales roughly linearly15 with dose, increasing from a value of ;0.4 Å at zero dose ~due mainly to
surface steps! to a value of ;15 Å at D ox5800 L.
IV. DISCUSSION
Smith and Ghidini2 have previously reported that
micrometer-sized conical structures could form at high temperatures ~.1000 °C! and high dose, provided the sample
temperature and O2 pressure were within a narrow transition
region between the active and passive oxidation regimes.
They postulated these structures to be Si cones covered with
a thin SiO2 layer, formed from a tiny oxide nucleus which
grew down the sides of the cone as the surrounding surface
was etched away.
We propose that the nanometer-sized conical islands observed in our experiments may in fact represent an early
stage of the larger structures reported by Smith and Ghidini.2
The smaller structures in our case are formed in a corresponding transition region between passive and active oxidation regimes, but at much lower temperature and O2 pressure.
In support of this connection, we note that the STM tunnel
current often became unstable at low tunnel voltages when
the tip was located over an island, but could generally be
stabilized by raising the tunnel voltage above a few volts.
This is consistent with the expected behavior if a thin layer
of SiO2 covers the island. We do note, however, that the
overall tip-to-width aspect ratio of the cones observed here is
substantially smaller than those observed by Smith and Ghidini. This may be because the intrinsic shape of the islands is
a strong function of oxidation temperature, because they
have not grown large enough to reach their limiting shape, or
even because the finite size of the STM tip causes the islands
to appear wider than they actually are. It is also possible,
775
J. V. Seiple and J. P. Pelz: Evolution of atomic-scale roughening
775
FIG. 5. Plot of rms surface roughness vs P ox for a 63 L dose at T s >600 °C.
The square symbol is initial rms roughness. STM scans used in this data
series were originally presented in Ref. 3.
FIG. 4. Temperature–pressure parameter space showing different oxidation
regimes. The dark solid line ~from Ref. 1! and the dark dashed line ~from
Ref. 2! represent a ‘‘critical line’’ which separates the passive and active
oxidation regimes. The dotted line shows the conditions which produce
maximum roughening during a 63 L O2 dose and is located well above the
critical line.
however, that the different shape is due to a completely different physical origin. More work is currently being done to
explore these issues.
The measurements reported here and recent STM measurements reported elsewhere3,4,8,16 also have interesting implications concerning the critical conditions for SiO2
growth,1,2 shown in Fig. 4. Earlier studies done mainly at
higher temperatures and with lower spatial resolution2 indicated a very sharp boundary between the active and passive
oxidation regimes, with large-scale surface roughening only
occurring over a narrow transition region a few degrees
wide.2 The more recent atomic resolution studies done at
lower temperatures3,4,8,16 all indicate that a great deal of surface etching can occur up to O2 pressures several orders of
magnitude above the commonly accepted ‘‘critical line’’ for
passive oxidation, at the same time that oxide clusters nucleate and grow.3,4,16 Consequently, the transition region in
which significant atomic-scale surface roughening occurs
may actually be much broader than would be expected from
the prior measurements. In fact, a recent transmission electron microscopy study has identified a roughening regime
under similar conditions on Si~111! which at P ox5131026
Torr is as wide as 125 °C.17 Although Ross, Gibson, and
Tweston17 did not quantify the roughness, they did attribute
its origin to simultaneous oxide nucleation and etching resulting in a micromasking process. As a practical matter for
device processing, it would be useful to know the range of
conditions which produce significant surface roughening on
Si~001! and the degree and type of roughness thus produced.
To explore surface roughening within this transition region, we have made measurements of how the rms surface
roughness depends on O2 pressure and temperature. This is
done keeping the total O2 dose constant to ensure that the
same amount of oxygen reacts with the surface at each pressure. Figure 5 shows results for a 63 L O2 dose at 600 °C. At
JVST A - Vacuum, Surfaces, and Films
zero dose the measured rms roughness is G>0.4 Å, due
mostly to regular steps on the clean vicinal surface. As the
O2 pressure is increased, we see that G first increases,
reaches a maximum near P ox5831028 Torr, and thereafter
decreases. For P ox.231027 Torr, STM images invariably
show a very smooth, ‘‘cloudy’’ topography, consistent with a
very thin, smooth layer of surface oxide. At these higher
pressures, a continuous surface oxide forms so quickly that
the surface has little chance to etch and roughen. Preliminary
measurements indicate that at 700 °C, a 63 L dose produced
maximum roughening at or above P ox5431026 Torr. These
two points can be used to draw an approximate line of
‘‘maximum roughening’’ for a 63 L dose, as shown in Fig. 4.
The position of this line and the peak value of the surface
roughness will of course depend on the oxygen dose. The
main point here is that substantial atomic-scale surface
roughening can occur for conditions well above the accepted
critical line. If one is designing a processing sequence for
devices which require atomically smooth interfaces, it is important to avoid these roughening conditions during transient
processing steps,5 i.e., as a substrate is heated or a chamber is
vented, etc.
Finally, we again note that Wurm et al.10 have recently
used LEEM to study oxidation on Si~001! under similar conditions as the present study ~P ox5531028 Torr, T s 5600 °C,
and D ox<100 L!, but did not report significant step pinning
or island nucleation. A possibility is that the lower lateral
resolution of the LEEM apparatus ~'15 nm! may have
missed the smaller islands we observed for D ox<100 L ~see
Fig. 2~a!#. Recent STM work18 by this group at these same
conditions did in fact show pinning similar to that observed
in this study, but at a much lower density. We also note that
Wurm et al.10 observed a sticking coefficient that was ;8
times smaller than ours.14 A possible cause for this difference
in the sticking coefficient may be different amounts of background water vapor ~which can greatly enhance silicon
oxidation17! or a difference in O2 ionization levels. Whatever
the case, it is important to note that the pinning density
should scale as the square of the sticking coefficient at a
given temperature, pressure, and dose as indicated in the
model from Ref. 3. This follows because the model predicts
that the oxide cluster nucleation rate varies quadratically
with the rate that oxygen actually reacts with the surface,
776
J. V. Seiple and J. P. Pelz: Evolution of atomic-scale roughening
which is proportional to the product s times P ox . Hence, the
lower sticking coefficient in the experiments of Wurm et al.
should result in a factor of ;64 lower pinning density and a
factor of ;8 less etching than observed in our study for
oxidation at the same temperature, pressure, and dose.
V. SUMMARY
When vicinal Si~001! surfaces are exposed to O2 at
T s >600 °C and P ox<631028 Torr, we observe the preferential retraction of B-type surface steps at low dose ~,50 L!,
but complete single A-domain formation is prevented by the
pinning of steps by nucleated oxide clusters. At higher doses,
these pinning sites produce three-dimensional conical etch
structures, which may be early stages of the micrometersized structures observed previously2 at higher temperature,
pressure, and dose. By measuring the evolution of these
structures with dose, we estimated surface etching rates and a
value of the O2 sticking coefficient s>0.0460.02 at
600 °C. We also point out that oxygen exposure at these
lower temperatures can cause significant atomic-scale surface roughening even up to O2 pressures several orders of
magnitude above the commonly accepted critical line for
SiO2 growth.
ACKNOWLEDGMENTS
The authors would like to acknowledge helpful discussions with G. Rubloff and F. Smith. This work was supported
by National Science Foundation Grant No. DMR93-57535,
PRF Grant No. 26647-G5, and The Ohio State University
Center for Materials Research.
J. Vac. Sci. Technol. A, Vol. 13, No. 3, May/Jun 1995
776
J. J. Lander and J. Morrison, J. Appl. Phys. 33, 2089 ~1962!.
F. W. Smith and G. Ghidini, J. Electrochem. Soc. 129, 1300 ~1982!.
3
J. V. Seiple and J. P. Pelz, Phys. Rev. Lett. 73, 999 ~1994!.
4
A. Feltz, U. Memmert, and R. J. Behm, Surf. Sci. 314, 34 ~1994!.
5
M. Offenberg, M. Liehr, and G. W. Rubloff, J. Vac. Sci. Technol. A 9,
1058 ~1991!.
6
J. V. Seiple and J. P. Pelz ~unpublished!.
7
P. Bedrossian and T. Klitsner, Phys. Rev. Lett. 68, 646 ~1992!.
8
J. Seiple, J. Pecquet, Z. Meng, and J. P. Pelz, J. Vac. Sci. Technol. A 11,
1649 ~1993!.
9
D. J. Chadi, Phys. Rev. Lett. 59, 1691 ~1987!.
10
K. Wurm et al., Phys. Rev. B 50, 1567 ~1994!.
11
P. B. Price, D. A. Vermilyea, and M. B. Webb, Acta Metall. 6, 524 ~1958!.
12
U. Memmert and M. L. Yu, Surf. Sci. Lett. 245, L185 ~1991!.
13
J. R. Engstrom, D. J. Bonser, M. M. Nelson, and T. Engel, Surf. Sci. 256,
317 ~1991!.
14
Wurm et al. report a reaction coefficient, r50.01060.005, where r[the
number of SiO molecules desorbed/the number of O2 molecules incident,
whereas we define s[the number of O2 molecules that etch the surface/
the number of O2 molecules incident. Thus r52s, so our sticking coefficient is ;8 times greater than theirs ~see Discussion Sec. IV!.
15
We do not fully understand why G does not scale as a higher power of the
dose, since the average height and base diameter of the islands both
clearly increase with dose. Part of the reason is related to the particular
number and size of the islands found in our experiments. If one models
the surface as a concentration n of conical islands each of height h and
base diameter D located on an otherwise flat surface, then
G2[^z 2 &2^z&25~pnh 2 D 2 /24) [12( p nD 2 /6!#. This turns out to be approximately independent of D when D>D 0 5(3/ p n) 1/2 . In our experiments, n'10 11 cm22 ~for dose.200 L! which implies D 0 >30 nm, as
compared with measured values of 15–30 nm. It is also possible that
subtle changes in measured island shape ~possibly due to the STM tip!
may be a factor.
16
F. Donig et al., J. Vac. Sci. Technol. B 11, 1955 ~1993!.
17
F. M. Ross, J. M. Gibson, and R. D. Twesten, Surf. Sci. 310, 243 ~1994!.
18
Y. Hong, K. Wurm, Y. Wei, and I. S. T. Tsong ~private communication!.
1
2